Anthropic declined to submit its Mythos 5.1 model to the UK's AI Safety Institute for pre-release testing, according to the Financial Times1,3. The decision marks a break from the voluntary testing arrangements that UK regulators had established with frontier AI developers.
The refusal has prompted fears within the British government that US AI labs are aligning with a broader US protectionist posture, according to sources cited by the Financial Times. The UK AISI, which conducts pre-deployment safety evaluations of advanced AI systems, was denied access to the model ahead of its release2.
The move comes during a period of intense activity for Anthropic. The company recently walked away from a roughly $6 billion acquisition of AI startup Decart after due diligence[2], and has locked in at least 14.8 GW of compute capacity in agreements since October, a buildout that could cost as much as $517 billion over the next decade[3].
ANALYSIS Anthropic's refusal to provide pre-release access to the UK AISI carries weight precisely because the company has positioned itself as a safety-focused lab. The decision to bypass a voluntary safety-testing regime, rather than a binding regulatory requirement, tests whether such arrangements can survive commercial and geopolitical pressures.
Separately, the Financial Times and other outlets reported that Anthropic researcher Jacob Coxon said he is leaving the AI industry over concerns that competition is pushing companies toward self-improving models that risk spiraling beyond human control.
ANALYSIS The timing of a safety researcher's departure alongside the AISI testing refusal concentrates attention on the gap between Anthropic's stated safety commitments and its operational decisions.